Towards Real-Time Prediction of Freezing of Gait in Patients With Parkinson’s Disease: Addressing the Class Imbalance Problem

Abstract

Freezing of gait (FoG) is a common motor symptom in patients with Parkinson’s disease (PD). FoG impairs gait initiation and walking and increases fall risk. Intelligent external cueing systems implementing FoG detection algorithms have been developed to help patients recover gait after freezing. However, predicting FoG before its occurrence enables preemptive cueing and may prevent FoG. Such prediction remains challenging given the relative infrequency of freezing compared to non-freezing events. In this study, we investigated the ability of individual and ensemble classifiers to predict FoG. We also studied the effect of the ADAptive SYNthetic (ADASYN) sampling algorithm and classification cost on classifier performance. Eighteen PD patients performed a series of daily walking tasks wearing accelerometers on their ankles, with nine experiencing FoG. The ensemble classifier formed by Support Vector Machines, K-Nearest Neighbors, and Multi-Layer Perceptron using bagging techniques demonstrated highest performance (F1 = 90.7) when synthetic FoG samples were added to the training set and class cost was set as twice that of normal gait. The model identified 97.4% of the events, with 66.7% being predicted. This study demonstrates our algorithm’s potential for accurate prediction of gait events and the provision of preventive cueing in spite of limited event frequency.

Early Detection of Freezing of Gait during Walking Using Inertial Measurement Unit and Plantar Pressure Distribution Data

“…In practice, the freeze identification model would perform very well as a FOG detection system, with a cue administered during the freeze if the Pre-FOG or transition states were missed. A similar analysis in [ 19 ] predicted 66.7% of the freeze episodes within 2 s of onset and detected 97.4% of the episodes between 2 s before and 4 s after FOG onset. These results were based on the number of FOG episodes, which may account for the higher performance compared to results presented in this paper, where results were based on decisions for each window.…”

“…PP-IMU features model sensitivity was 76.4%, indicating that approximately 24% of the target-class windows were missed by the model. Other FOG prediction research [ 13 ] reported higher sensitivity (93%), although as in [ 11 , 19 ], the performance metrics were calculated based on FOG episodes. Thus, the sensitivity results are not directly comparable to our window-based analysis.…”

“…Inertial measurement units (IMUs) on the lower limbs have been used for FOG classification [ 6 , 12 , 14 , 19 , 20 , 21 ]. In [ 14 ], the shank was the preferred sensor location for FOG episode prediction, and in [ 19 ], a model with ankle accelerometer data predicted 66.7% of the FOG episodes within 2 s prior to onset. To improve FOG identification performance, additional sensor types could be used.…”

Prediction of Freezing of Gait in Parkinson’s Disease Using Wearables and Machine Learning

“…Indeed, their usage, in combination with machine learning (ML) analysis, has recently smoothed the path for FOG prediction [ 29 , 30 , 31 , 32 , 33 , 34 ]. By examining several time and frequency-domain gait features, these studies have achieved the real-time detection of pre-FOG periods [ 29 , 31 , 32 , 35 ]. However, the reported performance of the ML analysis in the prediction of FOG is suboptimal in terms of accuracy, possibly reflecting the clinical heterogeneity of the cohorts under investigation.…”

Lower Limb Kinematics Trajectory Prediction Using Long Short-Term Memory Neural Networks

“…This enables the assistive device to adapt to changes in human gait, allowing smoother synchronisation with user intentions and minimising interruptions when the user changes their movement pattern (Elliott et al, 2014;Zhang et al, 2017;Ding et al, 2018;Zaroug et al, 2019). A known future trajectory might also monitor the risk of balance loss, tripping and falling, in which impending incidents can be remotely reported for early intervention (Begg and Kamruzzaman, 2006;Begg et al, 2007;Nait Aicha et al, 2018;Hemmatpour et al, 2019;Naghavi et al, 2019). Since 60 ms falls in the range of slow (60-120 ms) and fast (10-50 ms) twitch motor units (Winter, 2009), this would enable wearable devices such as IMUs to alert (e.g., by audio/visual signal) an elderly user about an imminent risk of tripping and potentially gives them a chance to adjust their gait accordingly.…”